Microsoft has released MAI-Transcribe-2-Streaming, its first real-time transcription model, on October 1, 2026. The model delivers live transcripts in 60 languages with automatic language detection, Microsoft AI announced.

The system produces its first transcript hypotheses just over 100 milliseconds after receiving audio, then revises them as more context arrives. Microsoft says words appear twice as fast as with its closest competitor, enabling live captions and voice agents that react mid-sentence.

On the Artificial Analysis benchmark, the model ranks first for accuracy among 38 models for both final and partial transcripts, with a 2.5 percent word error rate. Microsoft calls it Pareto-optimal on accuracy versus latency.

The model costs $0.54 per audio hour at an introductory price through the end of 2026. Microsoft also launched two text-to-speech models, MAI-Voice-2.1 and MAI-Voice-2.1-Flash, for building conversational voice agents.

The models are available through Microsoft Foundry, the MAI Playground, Vercel, and Azure Voice Live, with the voice models also on OpenRouter. Microsoft charges $22 per million characters for Voice 2.1 and $15 for the Flash variant. Both voice models support cloning from a few seconds of reference audio, with built-in consent checks to prevent misuse.